flowent 0.1.4 → 0.2.0

package/backend/tests/test_llm_providers.py

@@ -1,17 +1,29 @@
+import json
+
 import pytest
 
 from flowent.llm import (
     ChatMessage,
+    LLMStreamError,
     ProviderConnection,
     ProviderFormat,
     ReasoningEffort,
     build_litellm_request,
     chunk_delta_reasoning,
     complete_chat,
+    list_provider_models,
+    normalize_system_messages,
     stream_chat,
+    stream_chat_chunks,
 )
 
 
+def read_single_llm_request_diagnostic(tmp_path):
+    files = sorted((tmp_path / "logs" / "llm-requests").glob("llm-request-*.json"))
+    assert len(files) == 1
+    return json.loads(files[0].read_text())
+
+
 def test_supported_provider_formats_match_product_choices() -> None:
     assert [provider.value for provider in ProviderFormat] == [
         "openai",
@@ -47,6 +59,120 @@ def test_build_litellm_request_maps_provider_connection_to_completion_args() ->
     }
 
 
+def test_build_litellm_request_appends_default_api_version_to_base_url() -> None:
+    connection = ProviderConnection(
+        name="OpenAI Compatible",
+        provider=ProviderFormat.OPENAI,
+        model="gpt-5.1",
+        secret_reference="connection-primary",
+        base_url="https://example.test",
+    )
+
+    request = build_litellm_request(
+        connection, [ChatMessage(role="user", content="Draft a checklist.")]
+    )
+
+    assert request["api_base"] == "https://example.test/v1"
+
+
+def test_build_litellm_request_appends_anthropic_api_version_to_base_url() -> None:
+    connection = ProviderConnection(
+        name="Anthropic",
+        provider=ProviderFormat.ANTHROPIC,
+        model="claude-sonnet-4-5",
+        secret_reference="connection-primary",
+        base_url="https://example.test",
+    )
+
+    request = build_litellm_request(
+        connection, [ChatMessage(role="user", content="Draft a checklist.")]
+    )
+
+    assert request["api_base"] == "https://example.test/v1"
+
+
+def test_build_litellm_request_does_not_duplicate_existing_api_version() -> None:
+    connection = ProviderConnection(
+        name="OpenAI Compatible",
+        provider=ProviderFormat.OPENAI,
+        model="gpt-5.1",
+        secret_reference="connection-primary",
+        base_url="https://example.test/v1/",
+    )
+
+    request = build_litellm_request(
+        connection, [ChatMessage(role="user", content="Draft a checklist.")]
+    )
+
+    assert request["api_base"] == "https://example.test/v1"
+
+
+def test_build_litellm_request_preserves_custom_versioned_base_path() -> None:
+    connection = ProviderConnection(
+        name="OpenAI Compatible",
+        provider=ProviderFormat.OPENAI,
+        model="gpt-5.1",
+        secret_reference="connection-primary",
+        base_url="https://example.test/api/v3",
+    )
+
+    request = build_litellm_request(
+        connection, [ChatMessage(role="user", content="Draft a checklist.")]
+    )
+
+    assert request["api_base"] == "https://example.test/api/v3"
+
+
+def test_build_litellm_request_allows_base_url_version_opt_out() -> None:
+    connection = ProviderConnection(
+        name="OpenAI Compatible",
+        provider=ProviderFormat.OPENAI,
+        model="gpt-5.1",
+        secret_reference="connection-primary",
+        base_url="https://example.test#",
+    )
+
+    request = build_litellm_request(
+        connection, [ChatMessage(role="user", content="Draft a checklist.")]
+    )
+
+    assert request["api_base"] == "https://example.test"
+
+
+def test_build_litellm_request_appends_gemini_api_version_to_base_url() -> None:
+    connection = ProviderConnection(
+        name="Gemini",
+        provider=ProviderFormat.GEMINI,
+        model="gemini-3-pro-preview",
+        secret_reference="connection-primary",
+        base_url="https://example.test",
+    )
+
+    request = build_litellm_request(
+        connection, [ChatMessage(role="user", content="Draft a checklist.")]
+    )
+
+    assert request["api_base"] == "https://example.test/v1beta"
+
+
+def test_list_provider_models_uses_normalized_base_url() -> None:
+    captured_request: dict[str, object] = {}
+
+    def fake_model_lister(**request: object) -> list[str]:
+        captured_request.update(request)
+        return ["openai/gpt-5.1", "gpt-5.1-mini"]
+
+    models = list_provider_models(
+        provider=ProviderFormat.OPENAI,
+        secret_reference="connection-primary",
+        base_url="https://example.test",
+        model_lister=fake_model_lister,
+    )
+
+    assert captured_request["api_base"] == "https://example.test/v1"
+    assert models == ["gpt-5.1", "gpt-5.1-mini"]
+
+
 def test_build_litellm_request_omits_default_reasoning_effort() -> None:
     connection = ProviderConnection(
         name="Primary",
@@ -137,6 +263,63 @@ async def test_complete_chat_uses_injected_litellm_completion() -> None:
     assert answer == ChatMessage(role="assistant", content="Here is the checklist.")
 
 
+@pytest.mark.anyio
+async def test_development_mode_writes_completion_request_diagnostic_file(
+    tmp_path, monkeypatch
+) -> None:
+    monkeypatch.setenv("FLOWENT_DATA_DIR", str(tmp_path))
+    monkeypatch.setenv("DEBUG", "true")
+
+    async def fake_completion(**request: object) -> dict[str, object]:
+        return {
+            "choices": [
+                {
+                    "message": {
+                        "content": "Here is the checklist.",
+                        "role": "assistant",
+                    },
+                }
+            ]
+        }
+
+    connection = ProviderConnection(
+        name="Responses",
+        provider=ProviderFormat.OPENAI_RESPONSES,
+        model="gpt-5.1",
+        secret_reference="sk-request-secret",
+    )
+    messages = [ChatMessage(role="user", content="Create a checklist.")]
+    tools = [
+        {
+            "type": "function",
+            "function": {
+                "name": "create_checklist",
+                "description": "Create a checklist.",
+            },
+        }
+    ]
+
+    await complete_chat(
+        connection,
+        messages,
+        completion=fake_completion,
+        tools=tools,
+    )
+
+    diagnostic = read_single_llm_request_diagnostic(tmp_path)
+
+    assert diagnostic == {
+        "base_url": None,
+        "litellm_model": "openai/gpt-5.1",
+        "messages": [{"content": "Create a checklist.", "role": "user"}],
+        "model": "gpt-5.1",
+        "provider": "openai_responses",
+        "reasoning_effort": "default",
+        "stream": False,
+        "tools": tools,
+    }
+
+
 @pytest.mark.anyio
 async def test_stream_chat_uses_litellm_streaming() -> None:
     captured_request: dict[str, object] = {}
@@ -169,3 +352,197 @@ async def test_stream_chat_uses_litellm_streaming() -> None:
     assert captured_request["stream"] is True
     assert captured_request["model"] == "openai/gpt-5.1"
     assert chunks == ["Here is ", "the checklist."]
+
+
+@pytest.mark.anyio
+async def test_stream_chat_chunks_raises_when_responses_stream_fails(
+    fake_litellm_responses_transformer,
+) -> None:
+    async def fake_completion(**request: object) -> object:
+        async def chunks() -> object:
+            from litellm.completion_extras.litellm_responses_transformation.transformation import (
+                OpenAiResponsesToChatCompletionStreamIterator,
+            )
+
+            yield {"choices": [{"delta": {"content": "Partial answer."}}]}
+            yield OpenAiResponsesToChatCompletionStreamIterator.translate_responses_chunk_to_openai_stream(
+                {
+                    "response": {
+                        "error": {
+                            "code": "upstream_error",
+                            "message": "Upstream request failed",
+                        },
+                        "status": "failed",
+                    },
+                    "type": "response.failed",
+                }
+            )
+
+        return chunks()
+
+    connection = ProviderConnection(
+        name="Responses",
+        provider=ProviderFormat.OPENAI_RESPONSES,
+        model="gpt-5.1",
+        secret_reference="connection-responses",
+    )
+
+    with pytest.raises(LLMStreamError, match="Upstream request failed"):
+        [
+            chunk
+            async for chunk in stream_chat_chunks(
+                connection,
+                [ChatMessage(role="user", content="Create a checklist.")],
+                completion=fake_completion,
+            )
+        ]
+
+
+@pytest.mark.anyio
+async def test_development_mode_writes_one_streaming_request_diagnostic_file(
+    tmp_path, monkeypatch
+) -> None:
+    monkeypatch.setenv("FLOWENT_DATA_DIR", str(tmp_path))
+    monkeypatch.setenv("DEBUG", "true")
+
+    async def fake_completion(**request: object) -> object:
+        async def chunks() -> object:
+            yield {"choices": [{"delta": {"content": "Here is "}}]}
+            yield {"choices": [{"delta": {"content": "the checklist."}}]}
+
+        return chunks()
+
+    connection = ProviderConnection(
+        name="Responses",
+        provider=ProviderFormat.OPENAI_RESPONSES,
+        model="gpt-5.1",
+        secret_reference="sk-request-secret",
+    )
+
+    chunks = [
+        chunk
+        async for chunk in stream_chat(
+            connection,
+            [ChatMessage(role="user", content="Create a checklist.")],
+            completion=fake_completion,
+        )
+    ]
+    diagnostic = read_single_llm_request_diagnostic(tmp_path)
+
+    assert chunks == ["Here is ", "the checklist."]
+    assert diagnostic["stream"] is True
+
+
+@pytest.mark.anyio
+async def test_development_request_diagnostic_omits_api_key_and_secret_values(
+    tmp_path, monkeypatch
+) -> None:
+    monkeypatch.setenv("FLOWENT_DATA_DIR", str(tmp_path))
+    monkeypatch.setenv("DEBUG", "true")
+
+    async def fake_completion(**request: object) -> dict[str, object]:
+        return {
+            "choices": [
+                {
+                    "message": {
+                        "content": "Here is the checklist.",
+                        "role": "assistant",
+                    },
+                }
+            ]
+        }
+
+    connection = ProviderConnection(
+        name="Responses",
+        provider=ProviderFormat.OPENAI_RESPONSES,
+        model="gpt-5.1",
+        secret_reference="sk-provider-secret",
+    )
+    tools = [
+        {
+            "type": "function",
+            "function": {
+                "name": "create_checklist",
+                "description": "Uses api_key=sk-tool-secret when configured.",
+            },
+        }
+    ]
+
+    await complete_chat(
+        connection,
+        [ChatMessage(role="user", content="authorization=Bearer sk-message-secret")],
+        completion=fake_completion,
+        tools=tools,
+    )
+
+    rendered = next(
+        (tmp_path / "logs" / "llm-requests").glob("llm-request-*.json")
+    ).read_text()
+
+    assert "api_key" not in rendered
+    assert "sk-provider-secret" not in rendered
+    assert "sk-tool-secret" not in rendered
+    assert "sk-message-secret" not in rendered
+
+
+@pytest.mark.anyio
+async def test_non_development_mode_skips_request_diagnostic_file(
+    tmp_path, monkeypatch
+) -> None:
+    monkeypatch.setenv("FLOWENT_DATA_DIR", str(tmp_path))
+    monkeypatch.delenv("DEBUG", raising=False)
+
+    async def fake_completion(**request: object) -> dict[str, object]:
+        return {
+            "choices": [
+                {
+                    "message": {
+                        "content": "Here is the checklist.",
+                        "role": "assistant",
+                    },
+                }
+            ]
+        }
+
+    connection = ProviderConnection(
+        name="Responses",
+        provider=ProviderFormat.OPENAI_RESPONSES,
+        model="gpt-5.1",
+        secret_reference="sk-request-secret",
+    )
+
+    await complete_chat(
+        connection,
+        [ChatMessage(role="user", content="Create a checklist.")],
+        completion=fake_completion,
+    )
+
+    assert not (tmp_path / "logs" / "llm-requests").exists()
+
+
+def test_normalize_system_messages_keeps_multiple_system_messages_for_openai() -> None:
+    messages = [
+        {"role": "system", "content": "Base prompt."},
+        {"role": "system", "content": "Configured prompt."},
+        {"role": "user", "content": "Hello."},
+    ]
+
+    assert normalize_system_messages(messages, ProviderFormat.OPENAI) == messages
+
+
+def test_normalize_system_messages_converts_additional_system_messages_for_anthropic() -> (
+    None
+):
+    messages = [
+        {"role": "system", "content": "Base prompt."},
+        {"role": "system", "content": "Configured prompt."},
+        {"role": "system", "content": "Project prompt."},
+        {"role": "user", "content": "Hello."},
+    ]
+
+    assert normalize_system_messages(messages, ProviderFormat.ANTHROPIC) == [
+        {"role": "system", "content": "Base prompt."},
+        {"role": "user", "content": "Configured prompt."},
+        {"role": "user", "content": "Project prompt."},
+        {"role": "user", "content": "Hello."},
+    ]

package/backend/tests/test_logging.py

@@ -8,6 +8,7 @@ from flowent.logging import (
     configure_logging,
     ensure_logging_configured,
     redact_log_value,
+    sanitize_diagnostic_value,
 )
 
 
@@ -103,6 +104,35 @@ def test_logging_redacts_full_api_key_but_keeps_context(tmp_path, monkeypatch) -
     )
 
 
+def test_diagnostic_sanitizer_removes_secret_fields_and_values() -> None:
+    sanitized = sanitize_diagnostic_value(
+        {
+            "api_key": "sk-root-secret",
+            "messages": [
+                {
+                    "role": "user",
+                    "content": "authorization=Bearer sk-message-secret",
+                }
+            ],
+            "tools": [
+                {
+                    "function": {
+                        "name": "send_message",
+                        "description": "Needs api_key=sk-tool-secret.",
+                    }
+                }
+            ],
+        }
+    )
+
+    rendered = str(sanitized)
+
+    assert "api_key" not in rendered
+    assert "sk-root-secret" not in rendered
+    assert "sk-message-secret" not in rendered
+    assert "sk-tool-secret" not in rendered
+
+
 def test_direct_main_app_import_creates_data_log_file(tmp_path, monkeypatch) -> None:
     monkeypatch.setenv("FLOWENT_DATA_DIR", str(tmp_path))
     sys.modules.pop("flowent.main", None)

package/backend/tests/test_patch.py

@@ -0,0 +1,112 @@
+import pytest
+
+from flowent.patch import PatchError, affected_paths, apply_patch
+
+
+def test_apply_patch_applies_context_hunk_with_interleaved_changes(tmp_path) -> None:
+    target = tmp_path / "notes.txt"
+    target.write_text("start\nalpha\nmiddle\nbeta\nend\n")
+    patch = """*** Begin Patch
+*** Update File: notes.txt
+@@
+ start
+-alpha
++one
+ middle
+-beta
++two
+ end
+*** End Patch
+"""
+
+    result = apply_patch(patch, tmp_path)
+
+    assert result == {"files": [{"path": str(target), "status": "modified"}]}
+    assert target.read_text() == "start\none\nmiddle\ntwo\nend\n"
+
+
+def test_apply_patch_reports_context_mismatch(tmp_path) -> None:
+    target = tmp_path / "notes.txt"
+    target.write_text("start\nalpha\nend\n")
+    patch = """*** Begin Patch
+*** Update File: notes.txt
+@@
+ missing
+-alpha
++beta
+ end
+*** End Patch
+"""
+
+    with pytest.raises(PatchError, match=r"Patch context was not found\."):
+        apply_patch(patch, tmp_path)
+
+    assert target.read_text() == "start\nalpha\nend\n"
+
+
+def test_apply_patch_applies_multiple_hunks_in_order(tmp_path) -> None:
+    target = tmp_path / "notes.txt"
+    target.write_text("first\nsame\nend first\nsecond\nsame\nend second\n")
+    patch = """*** Begin Patch
+*** Update File: notes.txt
+@@
+ first
+-same
++one
+ end first
+@@
+ second
+-same
++two
+ end second
+*** End Patch
+"""
+
+    apply_patch(patch, tmp_path)
+
+    assert target.read_text() == "first\none\nend first\nsecond\ntwo\nend second\n"
+
+
+def test_apply_patch_keeps_simple_contiguous_replacement(tmp_path) -> None:
+    target = tmp_path / "notes.txt"
+    target.write_text("alpha\nbeta\n")
+    patch = """*** Begin Patch
+*** Update File: notes.txt
+@@
+-alpha
+-beta
++ready
+*** End Patch
+"""
+
+    apply_patch(patch, tmp_path)
+
+    assert target.read_text() == "ready\n"
+
+
+def test_affected_paths_reads_structured_patch_write_targets(tmp_path) -> None:
+    patch = """*** Begin Patch
+*** Update File: notes.txt
+@@
+-alpha
++beta
+*** Add File: created.txt
++hello
+*** Delete File: old.txt
+*** Update File: before.txt
+*** Move to: after.txt
+@@
+-before
++after
+*** End Patch
+"""
+
+    paths = affected_paths(patch, tmp_path)
+
+    assert paths == [
+        (tmp_path / "notes.txt").resolve(strict=False),
+        (tmp_path / "created.txt").resolve(strict=False),
+        (tmp_path / "old.txt").resolve(strict=False),
+        (tmp_path / "before.txt").resolve(strict=False),
+        (tmp_path / "after.txt").resolve(strict=False),
+    ]